Repository navigation
CollectiveX: ask InferenceX-app to ingest results that land on main / CollectiveX：结果合入 main 时通知 InferenceX-app 导入 - #4592
Conversation
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
| # Committed CollectiveX results on main are the dashboard's source of truth. Each push that changes | ||
| # them asks InferenceX-app to load that commit into its CollectiveX database. |
There was a problem hiding this comment.
🟡 nit (optional): The new workflow opens with a header comment explaining what it is for. AGENTS.md:56 bans these: "No header comments describing a file, topology, config, recipe, override, or section." Fix: remove lines 3-4, and put the source-of-truth rationale in the PR description or the workflow name.
Why this was flagged
In .github/workflows/collectivex-results-ingest.yml:3-4, a two-line comment between name: and on: describes the whole workflow. AGENTS.md:54 marks the Comments section as "mandatory, every file". AGENTS.md:56 forbids header comments describing a file or config. AGENTS.md:58 allows only technical explanation of a specific line or block. The comment explains no specific line, so the PR adds a convention violation the base branch does not have. merge-ingest.yml, the sibling workflow the PR cites as its model, has no such header.
Verification: Lines 3-4 of .github/workflows/collectivex-results-ingest.yml are a header comment that describes the whole workflow. AGENTS.md line 56 says "No header comments describing a file, topology, config, recipe, override, or section." The sibling .github/workflows/merge-ingest.yml has no such header. So the base branch has no violation here and this PR adds one.
A push to main that changes collectivex/results/ sends an ingest-collectivex repository_dispatch to InferenceX-app with the commit SHA, the same way merge-ingest hands benchmark results to the app. The app loads that commit's committed results into its CollectiveX database. CollectiveX：结果合入 main 时通知 InferenceX-app 导入 推送到 main 且修改了 collectivex/results/ 时，会向 InferenceX-app 发送带有提交 SHA 的 ingest-collectivex repository_dispatch，与 merge-ingest 将基准结果交给应用的方式相同。应用会将该提交中已提交的结果导入其 CollectiveX 数据库。
…mmit A dispatch sent before the app can receive it is dropped, so a manual run can resend any commit. The run name carries the commit, and a malformed SHA fails before dispatch. 可按指定提交手动重新发送结果导入。应用尚无法接收时发出的 dispatch 会被丢弃，因此可手动重新发送任意提交。 运行名称包含该提交；格式错误的 SHA 会在发送前失败。
1ce826b to
b4d0a38
Compare
Summary
Adds
.github/workflows/collectivex-results-ingest.yml. When a push tomainchangescollectivex/results/**, the workflow sends InferenceX-app aningest-collectivexrepository_dispatch with the commit SHA. This works the same waymerge-ingest.ymlhands benchmark results to the app. The app then loads that commit's committed results into its CollectiveX database (InferenceX-app#1532).workflow_dispatchtakes an optionalshaand resends that commit, for example one that landed before the app could receive it or whose ingest failed. A value that isn't a 40-character SHA fails before the dispatch is sent.permissions: {}. The only credential is the existingFRONTEND_PAT, as inmerge-ingest.yml.Validation
AI model disclosure
claude-opus-5-5, run through Claude Code): the workflow and this description.中文
摘要
新增
.github/workflows/collectivex-results-ingest.yml。推送到main且修改了collectivex/results/**时，该工作流向 InferenceX-app 发送带有提交 SHA 的ingest-collectivexrepository_dispatch，方式与merge-ingest.yml将基准结果交给应用相同。随后应用会把该提交中已提交的结果导入其 CollectiveX 数据库（InferenceX-app#1532）。workflow_dispatch接受可选的sha，用于重新发送该提交，例如应用尚无法接收时已合入的提交，或导入失败的提交。不是 40 位 SHA 的值会在发送 dispatch 前失败。permissions: {}。唯一的凭据是现有的FRONTEND_PAT，与merge-ingest.yml相同。验证